Advances in Virtual Powertrain Development

Simulation is critical for efficient product development, but proper tool selection and efficient data management are keys to success.

AVL Cruise M for real-time multidisciplinary system simulation offers a modeling domain for waste heat recovery systems. Organic rankine cycle for Energy Management Simulation is shown. (AVL)

Powertrain engineers have various tools at their disposal to develop new technologies — empirical modeling, component simulation, multi-system environments and physical testing chief among them. Determining which to employ often involves balancing tradeoffs between key criteria: speed, accuracy, flexibility, insight and repeatability.

Powertrain-development experts at the 2022 SAE COMVEC conference (left to right): AVL’s Thomas Howell, Eric Hruby of John Deere, Chang-Wook Lee from the Paccar Technical Center, Cummins’ Bob Tickel, and Paul Chambon of ORNL. (SAE/Ryan Gehm)

“We’ve got all these different techniques. How do we change the paradigm? How do we make it better, so we aren’t having to cope with these different tradeoffs to the same extent?” asked Thomas Howell, segment lead, conventional powertrain at AVL in the U.S., during an SAE COMVEC 2022 session.

AVL attempts to “cross-fertilize” the best approaches of hardware testing and the virtual simulation environment. One approach the company uses to minimize tradeoffs is Engineering Enhanced Models — specifically its AVL Cruise M model for multidisciplinary system simulation.

CONVERGE simulations of a hydrogen direct-injection engine. Shown is hydrogen-air mixing using LES (left) and RANS (right) turbulence models, run by Dr. Bifen Wu at Argonne National Laboratory. (Convergent Science)

“Typically, we will have a very detailed series of models, and one of them is combustion [in diesel engines]. The combustion model is time-consuming, and it’s difficult to get the simulation to work,” Howell explained. “So how can we make this faster? One of the approaches that we do is take the combustion model out and replace it with an empirical neural net model so that we can leverage that within the multi-system domain and get a faster solution.”

AVL leverages data amassed from testing engines over the course of decades. “We’ve got a lot of data,” Howell said. “We can use the data we’ve captured from all our past testing to really get an accurate model.”

Ansys Minerva is a simulation process and data management tool used by powertrain developers for data archiving and 3D visualization. (Ansys)

AVL can utilize this model for its hardware-in-the-loop environment. “Now we have a model which we’ve got great confidence in, it runs in real time, so we can run it against the actual controllers and can start doing calibration early on,” Howell said, using the example of an assessment of OBD (on-board diagnostics) monitors.

“We can do the calibration on OBD monitors when we identify, why is a certain monitor coming up when it shouldn’t be flagging under these environments?” he said. “We can adjust the calibration or even adjust the software very early on in the product cycle so that we don’t have the delays associated with that later on.”

Howell stressed that simulation is critical for efficient product development — but so too is proper tool selection. “We utilize it extensively for our design investigations to ensure that we have got the hardware decisions correct — but we need to ensure that the selection of the tool is correct,” he said. “It is key from a product-success standpoint that we understand the limitations of the tools.”

Cummins also recognizes the importance of multidisciplinary simulation. Bob Tickel, director of multidisciplinary simulation within research and technology at Cummins, referenced the modeFrontier design optimization platform, developed by Italian software company Esteco SpA, as an example during his 2022 COMVEC presentation.

“How do I pull all these different models together along with total cost of ownership (TCO), so when we sit down and meet with our customers we can say, what is it you’re looking for?” Tickel said. “Is it TCO? Is it initial cost? Reliability? Durability? We can walk through all those options and help them make decisions, because a single mine site might have multiple powertrains in their fleet, depending on the need.”

Simulating hydrogen combustion

Oak Ridge National Laboratory’s investment in high-performance computing enables the storage and management of exceptionally large volumes of simulation data. (Carlos Jones/ORNL, U.S. Dept. of Energy)

Destination Zero is Cummins’ initiative to drive CO2 emissions to zero by 2050, which inevitably requires batteries and hydrogen fuel cells to become dominant power sources. But in the current decade, the engine maker is still devoting a significant amount of engineering effort toward reducing NOx and CO2 emissions from the internal-combustion engine (ICE).

“And we will continue to do that; we don’t see the engine dying anytime too soon,” Tickel said. Hydrogen and other zero-carbon alternative fuels will play a major role in extending ICE life. He noted that Cummins is building its modeling and simulation toolset and capability in this area.

Converge software from Convergent Science is one solution that can conduct high-fidelity CFD studies and investigate different simulation phenomena, including hydrogen chemical kinetics, supersonic hydrogen jets, and fuel-air mixing. A team of researchers from Convergent Science, Argonne National Laboratory and Aramco Americas’ Detroit Research Center have been using the tool to accelerate the development of hydrogen propulsion systems. The team’s work is part of the Initiative for Modeling Propulsion and Carbon-neutral Transportation (IMPACT) Consortium.

John Deere employs simulation for aftertreatment optimization but acknowledges limitations of the modeling environment and digital twins, including a dearth of detailed knowledge and interactions in the chemical domain. (John Deere)

Accurate fuel chemistry is key to any hydrogen combustion simulation, the researchers said. Using 0D, 1D and 3D simulations, the team identified shortcomings in existing hydrogen kinetic mechanisms and employed a Monte-Carlo sampling approach to optimize the reaction rate coefficients. The optimized hydrogen mechanism led to significantly improved predictions of in-cylinder pressures and heat release rates, according to Convergent Science.

The company noted that capturing hydrogen injection and fuel-air mixing also is critical for accurate predictions of hydrogen combustion and emissions. The team tested different RANS and LES turbulence models to predict the mixture distribution in the combustion cylinder and various meshing strategies to capture the behavior of the hydrogen jet.

“Injecting hydrogen at very high pressures results in an underexpanded jet, meaning that the flow at the nozzle exit is supersonic,” Dr. Roberto Torelli, senior research scientist at Argonne, said in a statement. “This requires very fine meshes and very small time-steps to achieve an accurate solution, but this makes the simulation very expensive.”

By optimizing their meshing strategy, the team reportedly reduced simulation runtime by more than 50%, while maintaining accuracy. Convergent Science said the team is exploring “innovative methods” that could lead to another 80% runtime reduction to further improve the computational efficiency of hydrogen combustion simulations.

Simulation data management

Capturing and managing the large volume of data generated through simulation and modeling activities is a challenge, SAE COMVEC panelists agreed. Cummins is working with Ansys and its Minerva tool, which is powered by Aras, on enterprise-level simulation data management.

“We’ve been working on the tool, and we’ve pretty quickly gone from tens to hundreds and now approaching thousands of users,” Tickel said. “The point here isn’t necessarily just to archive our models and have them ready, but it’s about reuse. Let’s not rebuild models. Let’s not re-solve them. Let’s have them there ready to go, and then as technology and architectures change, we can pull down those models, combine them and use them in different ways.”

Ansys Minerva is available for both on-premises and cloud deployment. Companies must determine how they plan to proceed. “We’re doing lots of investigations around, is it on my local machine? Is it on on-premises computing resources? An HPC [high performance computing] system? The cloud? What are the tools that help connect these things together? And how do we make this all work?” Tickel posed. “So, we’re stepping through this cautiously. Like anything, it comes down to dollar and cents at the end of the day.”

These questions will be answered differently by different organizations. For example, Paul Chambon, senior R&D staff at Oak Ridge National Laboratory (ORNL), provided a “tongue in cheek” response: “We’re a national lab and we fortunately have the fastest computer in the world onsite. They want us to use it as much as possible, so now we’re looking at more synergies between the different pieces of software to make the most of all this processing power.”

The Frontier supercomputer at ORNL, which features 1.1 exaflops of performance, is reportedly the first to achieve the exascale level of computing performance, a threshold of a quintillion calculations per second. Testing and validation of the system was expected to be finished by the end of 2022, with full operation at the beginning of 2023.

AVL uses cloud computing from Amazon Web Services “for quite a bit of our stuff,” Howell said. “There are some things which we don’t want to put on the cloud, so we have some local servers, but a lot is in the cloud.”

John Deere also uses a balanced approach, said Eric Hruby, staff engineer for power unit development. “Our IT helps manage the interfacing, along with our simulation department, how to execute and utilize the various toolsets.”

Digital twins not a silver bullet

Physical prototypes will be obsolete within the next six years thanks to digital twin solutions. That’s the expectation of 67% of respondents to an independent survey of more than 2,000 professionals worldwide from various industries that was recently released by Altair.

While that seems unlikely in the powertrain-development realm, respondents’ belief that digital twin technology will reshape the way products are developed is certainly a safe bet. The survey found that 36% say the product-development timeline will speed up; 33% believe the need for physical prototypes will be reduced; and 28% say fewer simulations will be needed.

According to the survey, 69% of organizations are already leveraging digital twins; 71% of those businesses began investing in the technology in the past year. Areas most impacted by digital twin technology include more accurate risk assessments, faster time to market, and improved customer satisfaction (73%); reduced maintenance and warranty costs (62%); and more energy-efficient and/or less wasteful products or processes (73%).

“There’s little doubt data gleaned from using digital twins gives organizations a world of new insight,” James R. Scapa, founder and CEO of Altair, said in a statement. “But as this study confirms, we are just seeing the tip of the iceberg. As businesses realize the untapped benefits of taking digital twins to the next level, including the convergence of simulation technology, high-performance computing, and AI, the possibilities for revolutionizing industries, business processes and scientific research are endless.”

Despite the rosy outlook of this study, digital twins are not a silver bullet. Discussing John Deere’s use of modeling for aftertreatment optimization and thermal aging at the 2022 SAE COMVEC conference, Hruby addressed limitations of the modeling environment and digital twins. Models need to be grounded with physical data at various phases within the program, he stressed.

“What we tend to see is real-world failure modes are not typically known or characterized. If I’ve introduced a new component to my system, that’s the first time I’m experiencing it. I only know my fundamental interactions, but we know in reality there’s much more complex interactions that are present and that can cause a new failure mode,” Hruby said. “Those are gaps that have to be assessed through either additional physical testing or experience through the balance of simulation-physical test.”

Another challenge is that detailed knowledge and interactions in the chemical domain are not always known, unlike physical domains such as thermal properties and vibrations, Hruby continued. “[Chemical domain] tends to be much more difficult to characterize and to understand until you really get that experience into that environment – really what I’m talking about here are the chemical poisons,” he said, adding that this hurdle also applies to battery-electric solutions. “There’s a chemical and an electrical component, and there’s an interface between those two. There’s been great work and effort characterizing that, but there’s still many unknowns.”

Improvement opportunities

One area for improvement is cost. “When you have all these pieces of software, each have their own licenses and they can each do something great, you need them all, it ends up being very expensive overall,” said ORNL’s Chambon.

Cummins’ Tickel agreed, adding that better communication among the various simulation platforms is necessary. “Once you’ve invested, you don’t want to invest in two more tools or five more tools,” he said. “The codes can talk to each other – everybody says they can, but they don’t [always].”

Tickel said that FMI (Functional Mock-up Interface), an open standard for exchanging dynamical simulation models between different tools in a standardized format, and FMU (Functional Mock-up Unit), a file that contains a simulation model adhering to the FMI standard, help. “But we need more of that interchangeability and data exchange, making it that much easier,” he said. “That’s going to help us, because one location may be Abaqus, another might be Ansys or whatever else. More standards on how [tools] talk and work together is key.”